Efficient processing of spatial range queries on wireless broadcast streams

International Journal of Database Management Systems ( IJDMS ) Vol.6, No.1, February 2014
DOI : 10.5121/ijdms.2014.6103 29
EFFICIENT PROCESSING OF SPATIAL RANGE
QUERIES ON WIRELESS BROADCAST STREAMS
KwanHo In, HaRim Jung, Hee Yong Youn and Ung-Mo Kim
School of Information and Communication Engineering,
Sungkyunkwan University, Suwon, Korea
ABSTRACT
With advances in wireless networks and hand-held computing devices equipped with location sensing
capability (e.g., PDAs, laptops, and smart phones), a large number of location based services (LBSs) have
been successfully deployed. In LBSs, wireless broadcast is an efficient method to support the large number
of users. In wireless broadcast environment, existing research proposed to support range queries search,
may tune into unnecessary indexes or data object. This paper addresses the problem of processing range
queries on wireless broadcast streams. In order to support range queries efficiently, we propose a novel
indexing scheme called Distributed Space-Partitioning Index (DSPI). DSPI consists of hierarchical grids
that provide mobile clients with the global view as well as the local view of the broadcast data. The
algorithm for processing range queries based on DSPI is also proposed. Simulation experiments
demonstrate DSPI is superior to the existing index schemes.
KEYWORDS
Location dependent information services, continuous range queries, moving objects, index structures,
broadcast systems
1. INTRODUCTION
With advances in wireless technologies and the wide spread of mobile devices, the trend toward
pervasive computing has gained momentum. Since location (e.g., 2-dimensional coordinates) is
one of the most important properties in a pervasive computing environment, various location
dependent information services (LDISs) have emerged as one of the promising applications [3-4,
10-11]. In order to support LDISs, efficient processing of location dependent queries (LDQs),
which retrieve information based on the current location of mobile clients (MCs), is critical. One
of the essential classes of LDQs is the range query which finds out all data objects within a given
rectangular (or circular) region.
Wireless data broadcast is considered to be an effective way for disseminating location dependant
data (LDD) and supporting LDQs since it leverages computational capability of MCs, and thus
accommodates a huge number of MCs simultaneously. In this paper, we address the problem of
processing range queries (One of the essential classes of LDQs) in wireless data broadcast
environments.
Preliminaries. In wireless data broadcast, data (e.g., LDD) are periodically broadcasted. The
server periodically broadcasts data through public wireless channels. MCs then retrieve the data
to evaluate their queries on the broadcast stream, without sending any request to the server. An
important characteristic of wireless data broadcast is that data objects are sequentially delivered
on the air and available to MCs only when they are currently being broadcasted. In order to
retrieve the data necessary for query processing, MCs have to continuously listen to the broadcast

30
channel until the data arrive. This leads to vast energy consumption on MCs. Current mobile
devices usually support two operation modes: full operational mode called active mode and
energy conserving mode referred to as doze mode. MCs can conserve their energy if they can stay
in doze mode most of the time and switch to active mode only when the desired data object
arrives on the air.
Air indexing is commonly used to help MCs predict the arrival time of the desired data. The basic
idea is to interleave index information (e.g., arrival times of data) with data on the broadcast
stream. By first accessing the index information, MCs can selectively retrieve only the desired
data. However, the average waiting time of the MCs is increased because the insertion of index
information extends the broadcast cycle (e.g., the length of the wireless broadcast stream). In
wireless data broadcast, the following measures are commonly used as performance criteria [1]:
● Access Latency: The time elapsed from the moment an MC requests data to the moment all the
required data are retrieved by the MC. This corresponds to the waiting time of the MC.
●Tuning Time: The total amount of time spent by the MC in the active mode for actually listening
to the broadcast channel. The Tuning Time is commonly used for measuring the energy-
efficiency of wireless broadcast data access.
(a) Broadcast stream without an index.
(b) Broadcast stream with an index
Figure 1. Data retrieval without and with an index
Both Access Latency and Tuning Time are evaluated in terms of the number of buckets, where a
bucket is the smallest logical unit of wireless data broadcast. The broadcast stream consists of
index buckets and data buckets. Index buckets hold the index information, while data buckets

31
hold the data. A set of contiguous index buckets is referred to as an index segment and a set of
contiguous data buckets broadcasted between successive index segments is called a data segment.
Let us consider a server that periodically broadcasts dataset D, where D = {d1, d2, d3,…, d8},
where di(1 ≤ i ≤ 8) is a data object. Suppose that an MC wants to retrieve d8. Fig. 1(a) and Fig.
1(b) illustrate the process of data retrieval on the wireless broadcast stream without and with an
index, respectively. Let us assume that the MC tunes into the broadcast channel as depicted in the
figures.
Without an index, the MC has to continually listen to the broadcast channel until it retrieves d8 as
shown in Fig. 1(a). On the contrary, as illustrated in Fig. 1(b), the MC can stay in doze mode and
only wake up when d8 arrives after accessing index (e.g., i3 and i4). Note that each bucket
contains a temporal offset to the beginning of the next index segment to facilitate the access of
index. Although the MC reduces the Tuning Time with the help of index, it suffers from the
Access Latency due to the lengthened broadcast cycle. In order to support efficient query
processing in wireless data broadcast, the air index scheme has to minimize Tuning Time while
minimizing Access Latency by keeping the increase of the broadcast cycle minimal.
Contributions. In this paper, we propose a novel air index structure, called Distributed Space
Partitioning Index (DSPI), for supporting range queries in wireless data broadcast environments.
DSPI consists of a hierarchy of grids, each of which uniformly partitions the data space with
different granularity. Additionally, DSPI has a linear, and yet distributed structure suitable for
wireless data broadcast. The technical contributions of the DSPI are summarized as follows:
● Providing efficient guidance for MCs to selectively retrieve only the required data objects by
letting the MCs traverse a hierarchy of grids.
● Enabling MCs to quickly process range queries by reducing the size of index.
● Having a linear, and yet distributed structure suitable for sequential data delivery in wireless
data broadcast while preserving the spatial locality of data objects.
Organization. The rest of the paper is organized as follows. In Section 2, related work to our
study is presented. Section 3 presents the proposed index scheme, namely DSPI. In Section 4,
experiments are conducted to show the efficiency of DSPI. Finally, we conclude this paper in
Section 5.
2. RELATED WORKS
In mobile computing environments, wireless broadcasting has been widely adopted for delivering
information to a large number of users due to its scalability benefit [12-15]. In a wireless
broadcasting system, the server periodically broadcasts data objects through the wireless
broadcasting channel in a pre-scheduled sequential order. When each mobile device receives a
query from a user, it tunes into the channel and processes the query on the broadcast stream.
Air index interleaving [16] is commonly used for reducing the tuning time at the expense of the
increased access time. The basic idea is to interleave an index information with data objects on
the broadcast stream. By first examining the index information, the mobile client can get the
arrival times of the relevant data objects for the given query. A wireless data broadcast, the
concept of air indexing has received much attention and was first discussed in [1]. In [1], the
authors proposed the (1, m) indexing and the distributed indexing schemes. Both schemes apply
an index tree (e.g., B+
-tree), where each node stores the ids and arrival times of its children. In
order to start query processing, an MC should wait until the root node of the index tree arrives.

32
Fig. 2 shows an example of the index tree for a broadcast stream. Circles represent index nodes
and the rectangles in the bottom represent data objects.
Figure 2. Example of the index tree
In the (1, m) indexing, the entire index is replicated m times and inserted into every 1/m fraction
of the broadcast stream (See Fig. 3). By replicating the entire index for m times, the waiting time
for reaching the root node can be reduced. This, however, incurs a huge amount of Access
Latency since the duplication of the entire index lengthens the broadcast cycle.
Figure 3. Example of (1, m) indexing
In the distributed indexing, an index is partially replicated and only the relevant portion of the
index is inserted in front of the data segment which immediately follows it. Since the distributed
indexing scheme replicates only the upper level of the index tree (Replicated Part in Fig. 2), the
size of broadcast stream is small compared with the that of (1, m) indexing. As a result,
distributed indexing shows better performance than the (1, m) indexing in terms of Access
Latency. Both (1, m) indexing and distributed indexing focus only on the id-based data retrieval.
That is, data objects are retrieved based on their identifiers. To process LDQs on the wireless data
broadcast stream, an indexing scheme should support location-based data retrieval. Recently,
several index structures for supporting LDQs in wireless data broadcast have been proposed [4, 5,
7, 8]. Among them, we give a brief overview of the latest air indexing schemes, namely Hilbert
Curve Index (HCI ) [7] and Distributed Spatial Index (DSI) [8].
To fit the sequential nature of wireless data broadcast, HCI and DSI utilize the Hilbert curve
(HC), one of the well-known space-filling curves. The Hilbert curve is a continuous path which
visits all cells in a multi-dimensional grid exactly once without crossing itself. Fig. 4(a) and 4(b)

33
show an example of Hilbert curves of order 1 and 2, for 21×21 grid and 22×22 grid respectively.
Numerical values in the figures represent the visiting order of the curves.
(a) Order 1 (b) Order 2
Figure 4. Hilbert curves of order 1 and 2.
Since the Hilbert curve can efficiently map 2-dimensional space to 1-dimensional space without
losing the locality of the underlying data, it facilitates the linear scan of the 2-dimensional data
delivered through the wireless broadcast channel. Specifically, the location (e.g., coordinates) of
each data object is approximated by a unique integer called HC value. This is done by partitioning
the data space into 2λ
×2λ
grid cells, so that each cell contains only one data object. Here, λ is the
order of Hilbert curve. Then each cell is assigned a HC value according to the visiting order of
Hilbert curve. Finally, the location of the data object inside each cell is approximated by the
corresponding HC value. Fig. 5(a) and 5(b) illustrate a sample dataset containing 32 data objects
and their approximated locations respectively. For example, the locations of data objects d1 and
d2 are approximated by 2 and 5.
In HCI and DSI, data objects are broadcast in the increasing order of their HC values. HCI
constructs a B+
-tree, where leaf nodes contain the HC values of all the data objects together with
corresponding pointers. Each data object is indexed and identified by its HC value (e.g.,
approximated location), and the associated pointer indicates the arrival time of the data object. To
reduce the waiting time for reaching the root node of the index, HCI utilizes the (1, m) indexing
scheme.
DSI, a table-based indexing scheme, organizes the index structure in a fully distributed fashion.
The basic idea of DSI is to split the broadcast stream into equal-sized frames, each of which
contains (i) a number of data objects and (ii) index table. An index table consists of logrNF entries,
where r is the exponential base and NF is the number of frames in a broadcast cycle. The ith entry
of any index table is represented as a < HCi, ptri > tuple. Here, HCi is the smallest HC value of
the data objects within the frame pointed by ptri, and ptri is the pointer which indicates the arrival
time of the next ri
th frame. In this manner, DSI enables each frame to maintain the indexing
information of the data objects to be broadcast.

34
(a)Dataset (b)HC approximation (λ=3)
Figure 5. Dataset and Hilbert curve approximation
With both HCI and DSI, an MC has to retrieve data objects more than necessary to process a
range query, because the locations of data objects are approximated by the HC values. For
example, consider the processing of the range query RQ (shaded rectangle) in Fig. 5 (b). With the
HCI, the MC should retrieve all data objects, whose HC value ∈{17, 23, 24, 28, 30, 32, 33, 36,
39, 40, 46}. However, only the data objects, whose HC value ∈ {28, 33}, can be the result of
RQ. Although DSI reduces the number of data objects being retrieved, it cannot completely
eliminate the unnecessary retrieval of data objects. This increases Tuning Time of the MC.
Furthermore, not only the distribution of LDD in the original space but also the number of LDD
may greatly deviate the performance of both HCI and DSI in terms of Access Latency as well as
Tuning Time.
The bucket is the basic unit of wireless broadcasting, and the broadcast stream consists of index
buckets and data buckets. Index buckets hold index information, whereas data buckets hold data.
We assume that each bucket has the same capacity, irrespective of its type (i.e., index bucket or
data bucket). In order to make all buckets self-identifying, each bucket contains the following
header information: bucket id, bucket type, upcoming appearance time of the next index bucket in
the broadcast stream, and start time of the next broadcast stream. Note that the bucket id is
determined by the offset of the corresponding bucket from the start time of the current broadcast
stream [17]. In the remainder of this paper, we use the number of buckets to measure access time
and tuning time. The number of buckets can be translated into time values without loss of
generality because all buckets are assumed to have the same capacity and the bandwidth of the
broadcasting channel is fixed.
3. DISTRIBUTED SPACE-PARTITIONING INDEX (DSPI)
In this section, we propose Distributed Space-Partitioning Index (DSPI) that enables MCs to
selectively retrieve only the required data objects (e.g., LDD). The proposed indexing scheme
significantly reduces Tuning Time and Access Latency compared to the previous indexing
schemes in [3, 4].

35
3.1. Index Structure
DSPI consists of nG hierarchical levels of grids, where nG is the number of grids. Each level of
grids partitions the data space with different granularity into grid cells of the same size. Fig. 6
shows the basic idea of DSPI, when nG = 2. For simplicity, we assume that nG = 2 in the rest of
the paper. However, extension to nG > 2 is straight forward.
Figure 6. Hierarchical grids
The lowest level of the grids, referred to as the leaf grid, indexes the underlying dataset, whereas
the upper level of the grids, called the directory grid, indexes the leaf grid. For the linear
placement of the grid cells as well as data objects on the air, we use the Hilbert curve since it
preserves the spatial locality better than other curves. Let DS be a two-dimensional data space.
For simplicity, we assume that DS is the unit square (0, 1)2
in our discussion.
Definition 1: (Leaf grid). The leaf grid is a grid that uniformly partitions the DS into 2γL
×2γL
grid
cells, where γL (≥ 1) is the granularity factor of the leaf grid. We call a grid cell in the leaf grid a
leaf cell.
Figure 7. Leaf grid (γL=2)
The leaf grid has the finest granularity among the grids in DSPI. It provides MCs with local
view of the underlying dataset D. The extent of a leaf cell on every dimension is 1/2γL
, so that the
leaf cell lc(i,j) at column i and row j (starting from the low-left corner of the leaf grid) maintains
the detailed information of all data objects with x-coordinate in the range (i×1/2γL
, (i+1)×1/2γL
)
and y-coordinate in the range (j×1/2γL
, (j+1)×1/2γL
). Specifically, each leaf cell maintains the
following information:
●Identifier: A leaf cell is assigned a HC value as an identifier according to the occurring order
on the Hilbert curve of order λ = γL. The identifier determines the broadcast order of the leaf cell.

36
●Object index table: An object index table stores identifiers, locations and arrival times of all
data objects which belong to the leaf cell. It consists of a set of tuples in the form of <id, (x, y),
ptr>.
●Directory cell pointer dc_ptr: dc_ptr indicates the arrival time of the next directory cell.
The leaf grid enables MCs to retrieve only the required data objects for range query processing by
using the 2-dimensional coordinates, instead of approximated coordinates, of data objects. Figure
7 illustrates the structure of the leaf grid based on the sample dataset in Fig. 5(a). In the broadcast
stream, all the leaf cells are sorted according to their identifiers (e.g., HC values), and each leaf
cell is placed in front of the relevant data segment containing data objects belonging to the leaf
cell.
Definition 2: (Directory grid). The directory grid is a grid that uniformly partitions the DS into
2γD
×2γD
grid cells, where γD (1 ≤ γD < γL) is the granularity factor of the directory grid. We call a
grid cell in a directory grid a directory cell. In particular, the directory grid which has the coarsest
granularity is called the root grid.
Figure 8. Directory grid (γD =1)
Since we assume nG = 2, hereafter we use directory grid and root grid interchangeably. Directory
grid maintains information of the lower level grid, namely, the leaf grid (Note that we assume nG
= 2) as well as the global view of D. The extent of each directory cell on every dimension is 1/2γD
.
The directory cell dc(i,j) maintains the information of its child cells, i.e., all leaf cells inclosed by
the range (i×1/2γD
, (i+1)×1/2γD
)× (j× 1/2γD
, (j+1) × 1/2γD
). For example, the directory cell dc(0,0) in
Fig. 8 contains the information of the child cells lc(0,0), lc(1,0), lc(0,1) and lc(1,1) in Fig. 7. Each
directory cell includes the following information:
●Identifier: A directory cell is assigned an HC value as an identifier according to the
occurring order on the Hilbert curve of order λ = γD. The identifier determines the broadcast
order of the directory cell.
●Child index table: A child index table stores the identifier, the minimum bounding rectangle
(MBR) and the arrival time of each child cell. The child index table consists of a number of
tuples in the form of <id, MBR, ptr>. The MBR is a rectangle of minimum area that fully
encloses all the data objects belonging to the corresponding child cell.
●Directory cell pointer dc_ptr: dc_ptr indicates the arrival time of the next directory cell.
The directory grid enables MCs to selectively read only the necessary portion of the leaf grid, by
examining the MBRs of its child cells. Fig. 8 illustrates the structure of the directory grid based

37
on the leaf grid in Fig. 7. All the directory cells are sorted according to their identifiers and each
of them is placed in front of its child cells on the broadcast stream. Fig. 9 illustrates the wireless
broadcast stream generated from the hierarchical grids shown in Fig. 7 and 8.
Figure 9. Broadcast stream generated from the hierarchical grids
3.2. Range Query Processing
A range query retrieves all data objects within a given rectangular (or circular) region. More
formally, given a set of 2-dimensional dataset D= {d1, d2, …, d|D|}and a query region q, the result
of range query is R(q)= {d∈D | d is within q}. In our discussion, the query region q is
represented by a rectangle.
To process a range query using DSPI, an MC first detects a collection of directory cells which
overlap with the query region q, and thereafter it retrieves the qualified data objects by traversing
the directory cells and their child cells on the broadcast stream.
In particular, given a set of overlapped directory cells, an MC maintains a list L of their identifiers
(ids). Note that given the mapping function of the Hilbert curve, it is easy for an MC to determine
the ids (HC values) of the overlapped directory cells. Then, the MC performs initial probe, i.e.,
tuning into the broadcast channel to find out when the next directory cell arrives on the broadcast
stream. After the initial probe, the MC continues to perform the following steps until L is empty:
1. The MC sequentially accesses a number of directory cells until it reaches the closest
directory cell dc on the broadcast stream whose id∈L. This is done by following dc_ptr in
each accessed directory cell. Then, the MC removes the id of dc from L.
2. Using the child index table in dc, the MC determines (i) the child cells whose MBRs
intersect with the query region q and (ii) when to tune into the broadcast channel in order to
access them.
3. The MC selectively accesses each qualified child cell. For each qualified child cell lc, the
MC determines (i) the data objects which are within the query region q and (ii) the arrival
times of them by using the object index table in lc. Then, the MC selectively retrieves the
qualified data objects.

38
Example. Consider the range query with its query region q (shaded rectangle) either in Fig. 7 or 8.
As the first step, the MC determines a set of overlapped directory cells, namely dc(0,1) and dc(1,1),
from the directory grid in Fig. 8 and inserts their ids (dc1 and dc2) into the list L. Then, the MC
tunes into the broadcast channel to find out when the next directory cell will arrive on the
broadcast stream.
Without loss of generality, we assume that the MC first tunes into the broadcast channel at the
time (or position) of dc1, which contains the information of its child cells (lc4, … , lc7) as well as
their MBRs. Since dc1∈L, the MC removes dc1 from L and determines the child cells whose
MBRs intersect with the query region q. Because only the MBR of lc7 overlaps with the q as
shown in Fig. 8, the MC accesses only lc7 and selectively retrieves the qualified data object,
namely d14, by utilizing the object index table in lc7. The MC then accesses to the next closest
directory cell on the broadcast stream whose id∈L, namely dc2, by using the dc_ptr in dc1. Then,
the MC removes dc2 from L. Since MBRs of lc8 and lc11 overlap with the query region q (See Fig.
8), the MC accesses both lc8 and lc11, and then retrieves d17. In this way, the MC completes the
processing of range query. Algorithm 1 presents the detailed algorithm for processing range
queries based on DSPI.
4. PERFORMANCE EVALUATION
We evaluated the performance of DSPI by comparing its Access Latency and Tuning Time with
those of HCI and DSI. For the implementation of the HCI, the (1, m) indexing scheme was
utilized. We used the optimal m (=√(DATAsize/INDEXsize)) to minimize Access Latency, where
DATAsize and INDEXsize are the sizes of whole data objects and the global index in terms of the
number of buckets [1]. On the other hand, for the DSI, the exponential base r was set to 2 and the
number of data objects within each frame was determined so that the index table associated with a
frame could fit into one bucket [8]. For the configuration of DSPI, we set nG = 2. In addition, the
granularity factor γL of the Leaf Grid was set to 4, and the granularity factor γD of Directory Grid
was set to 3. The system model, which consists of a broadcast server, MCs and a broadcast
channel, was implemented in the Java language.

39
We conducted performance analysis on two datasets (See Fig. 10): UNIFORM dataset and REAL
dataset. In the UNIFORM dataset, 6,000 data objects are uniformly generated in a square
Euclidian space. The REAL dataset contains 5,848 actual cities and villages of Greece, which is
extracted from the dataset available form [9]. In the experiment, 10,000 queries were issued. We
defined the ratio of average side length of a query rectangle to that of the whole data space,
referred to as RectSideRatio. We used the number of bytes instead of the number of buckets to
evaluate Access Latency and Tuning Time because the bucket size is varied in the experiment [8].
The number of bytes can be translated into the time values without loss of generality, since the
bandwidth of wireless channel is fixed. The size of a data object is 1024 bytes, and 16 bytes are
used for representing the 2-dimensional coordinates of a data object (8 bytes for each coordinate).
16 bytes are also used for representing the HC value. We set the pointer size to 2 bytes. Table 1
illustrates parameter settings used in the experiments. Since the number of MCs does not affect
the system performance [7], we considered only one MC in the performance evaluation.
(a)UNIFORM (b)REAL
Figure 10. Datasets for performance evaluation
Table 1. Parameter Settings
4.1. Evaluation of Access Latency
Fig. 11 shows the Access Latency performance when we fixed RectSideRatio to 0.1 and varied
bucket capacity from 26
to 29
bytes. For both datasets, namely UNIFORM (Fig. 11
(a)) and REAL (Fig. 11(b)), DSPI is superior to the other two indexing schemes (HCI, DSI).
This is due to the fact that the overall length of the broadcast cycle in DSPI is shorter than those
in the other two indices. Bucket capacity of UNIFORM dataset, DSPI decrease 51.7% and 30.6%
in terms of the Access Latency over HCI and DSI. Bucket capacity of REAL dataset, DSPI
decrease 51.1% and 16.6% in terms of the Access Latency over HCI and DSI. In other words, the
index size of HCI and DSI is much larger than that of DSPI. In HCI, the whole index is
interleaved m times with data objects, while in DSI, every frame has to maintain lots of
information (e.g., HC values of logrnF frames and pointers to those frames) when a large number
of data objects are concerned. Furthermore, in DSI, redundant information increases as the

40
number of data objects increases. Note that the Access Latency of DSPI, for the bucket size = 29
,
increases more rapidly than that of the others. This happens because the size of index segment,
i.e., a number of contiguous index buckets, is much larger than the size of the directory (or leaf)
cell. The Access Latency performance of DSPI tends to be affected by the bucket capacity due to
its small index size.
(a)UNIFORM dataset (b) REAL dataset
Figure 11. Access Latency versus bucket capacity
Figure 12. Access Latency versus RectSideRatio
For a more comprehensive evaluation, we fixed bucket capacity to 27
bytes and obtained the
simulation results by varying RectSideRatio. Obviously, as depicted in Fig. 12, Access Latency of
all the indices for both datasets gradually increases as the RectSideRatio increases. RectSideRatio
of UNIFORM dataset, DSPI decrease 56.3% and 31.3% in terms of the Access Latency over HCI
and DSI. RectSideRatio of REAL dataset, DSPI decrease 55.6% and 30.4% in terms of the Access
Latency over HCI and DSI. However, DSPI shows better performance than the other indices.

41
4.1. Evaluations of Tuning Time
Note that the Tuning Time affects the energy consumption on MCs. Here, we evaluated the
Tuning Time performance. By fixing the RectSideRatio to 0.1 and varying bucket capacity from
26
to 29
bytes, Fig. 13 shows the Tuning Time performance of the three indexing schemes. Fig. 14
plots the Tuning Time performance under a fixed bucket capacity and varied RectSideRatio.
Figure 13. Tuning Time versus bucket capacity
Figure 14. Tuning Time versus RecsideRatio
As shown in both figures, DSPI outperforms HCI and DSI. Bucket capacity of UNIFORM dataset,
DSPI decrease 55.3% and 17.1% in terms of the Tuning time over HCI and DSI. Bucket capacity
of REAL dataset, DSPI decrease 57.4% and 24.1% in terms of the Tuning time over HCI and DSI.
RectSideRatio of UNIFORM dataset, DSPI decrease 65.1% and 16.5% in terms of the Tuning
time over HCI and DSI. RectSideRatio of REAL dataset, DSPI decrease 65.1% and 29.9% in
terms of the Tuning time over HCI and DSI. This is because DSPI allows MCs to avoid reading

42
unnecessary data objects by offering the global view (MBRs) as well as the local view
(coordinates of the data objects) of the underlying dataset. As mentioned in Section 2, both HCI
and DSI are constructed based on the HC values of the underlying data objects instead of their
exact coordinates. As a result, MCs have to listen to broadcast channel more than necessary. This
incurs a huge amount of Tuning Time. Furthermore, as the order of Hilbert curve increases (due
to the skewed distribution or a large number of data objects), spatial locality of the data objects in
2-dimensional space cannot be preserved in the 1-dimensional linear space. Therefore, MCs have
to read much more data objects than necessary.
5. CONCLUSION
This paper addresses the problem of processing range queries on wireless broadcast streams. In
order to support range queries efficiently, this paper proposed a novel index structure, called
Distributed Space Partitioning Index, for processing range queries in wireless broadcast
environments. DSPI utilizes the notion of hierarchical grids in order to allow MCs to selectively
retrieve the required data in an efficient way in terms of Access Latency as well as Tuning Time.
As demonstrated in the simulation study, DSPI outperforms the existing index schemes (e.g., HCI,
DSI).
ACKNOWLEDGEMENTS
This research was supported by Basic Science Research Program through the National Research
Foundation of Korea (NRF) funded by the Ministry of Education (NRF-2013R1A1A2008578)
&& This research was funded by the MSIP (Ministry of Science, ICT & Future Planning), Korea
in the ICT R&D Program 2013
REFERENCES
[1] T.Imielinski, S.Viswanathan and B.R.Bardrinath, Data on air: Organization and access, IEEE
Transaction on Knowledge and Data Engineering (TKDE), 9(3): pp. 353- 372, 1997.
[2] A. Datta, D. VanderMeer, A. Celik and V. Kumar, Broadcast protocols to support efficient retrieval
from databases by mobile users, ACM Transactions on Database Systems, 24(1): pp. 1-79, March
1999.
[3] C. Gostman and M. Linderbaum, On the metric properties of discrete space-filling curves, IEEE
Transactions on Image Processing, 5(5): pp. 794-797, May 1996.
[4] J. Xu, B. Zheng, W.C. Lee and D.L. Lee, Energy efficient index for querying location-dependent data
in mobile broadcast environments, Proc. 19th IEEE Conf. ICDE 03, 5(5): Bangalore, India, March
2003.
[5] B. Zheng, J. Xu, W.C. Lee and D.L. Lee, Energy conserving air indexes for nearest neighbor search,
Proc. 9th Conf. EDBT 04, 5(5): Heraklion, Crete, Greece, March 2004.
[6] J. Xu, W.C. Lee, X. Tang, Q. Gao and S. Li, Exponential index: a parameterized distributed indexing
scheme for data on air, Proc. 2nd ACM Conf. MobiSys 04, Boston, Massachusetts, USA: pp. 153-
164, June 2004.
[7] B. Zheng, W.C. Lee and D. Lee, Spatial queries in wireless broadcast systems, Wireless Network,
10(6): pp. 723-736, December, 2004.
[8] W. Lee and B. Zheng, DSI: A Fully Distributed Spatial Index for Location-based Wireless Broadcast
Services, Proc. 25th IEEE Conf. ICDCS 05, Columbus, Ohio, USA, pp. 349-358, June 2005.
[9] Spatial Datasets. available at http://www.rtreeportal.org.spatial.html.
[10] S. Ilarri, E. Mena, and A. Illarramendi, Location-Dependent Query Processing: Where We Are and
Where We Are Heading, ACM Computing Surveys, 42 (3) (2010), pp. 1-73.
[11] B. Gedik, and L. Liu, Mobieyes: A Distributed Location Monitoring Service Using Moving Location
Queries, IEEE Transactions on Mobile Computing, 5 (10) (2006), pp. 1384-1402.
[12] Y. D. Chung, An indexing scheme for energy-efficient processing of content-based retrieval queries
on a wireless data stream, Information Sciences 177 (2) (2007), pp. 525-542.

43
[13] A. B. Waluyo, W. Rahayu, D. Taniar and B. Srinivasan, A novel structure and access mechanism for
mobile data broadcast in digital ecosystems, IEEE Transactions on Industrial Electronics, 58 (6)
(2011), pp. 2173-2182.
[14] J. Zhong, W. Wu, Y. Shi and X. Gao, Energy-efficient tree-based indexing schemes for information
retrieval in wireless data broadcast, Lecture Notes in Computer Science 6588 (2011), pp. 335-351.
[15] Y. Kwon, H. Jung, Y. D. Chung, Monitoring continuous k-nearest neighbor queries in the hybrid
wireless network, Journal of Zhejiang University SCIENCE, 12(3), 2011, pp. 213-220.
[16] H. Jung, B. K. Cho, Y. D. Chung and L. Liu, On processing location based top-k queries in the
wireless broadcasting system, in Proceedings of ACM Symposium on Applied Computing, 2010, pp.
585-591.
[17] H. Jung, Y. D. Chung and L. Liu, Processing generalized k-nearest neighbor queries on a wireless
broadcast stream, Information Sciences, 188, 2012, pp. 67-79.
[18] A. B. Waluyo, W. Rahayu, D. Taniar and B. Srinivasan, A novel structure and access mechanism for
mobile data broadcast in digital ecosystems, IEEE Transactions on Industrial Electronics, 58 (6)
(2011), pp. 2173-2182.
[19] K. Mouratidis, and S. Bakiras, Continuous Monitoring of Spatial Queries in Wireless Broadcast
Environments, IEEE Transactions on Mobile Computing, 8 (10) (2009), pp. 1297-1311.
AUTHORS
Kwanho In received his B.S. degree in Computer Science from Dongguk University,
Gyeongju, Korea, in 2009. He received the M.S. degree in information and
communication engineering from Sungkyunkwan University, Suwon, Korea. His
research interests include database system, spatial data management in
mobile/pervasive environments.
HaRim Jung received his B.S. degree in Computer Science from Kwangwoon
University, Seoul, Korea, in 2004. He received his M.S. and Ph.D. degrees in Computer
Science and Engineering from Korea University, Seoul, Korea, in 2007 and 2012,
respectively. Currently, he is a research fellow at the School of Information and
Communication Engineering, Sungkyunkwan University, Suwon, Korea. His research
interests include location-based services, spatial data management in mobile / pervasive
environments and spatial big data management.
Hee Yong Youn received the B.S. and M.S. degrees in electrical engineering from
Seoul National University, Seoul, Korea, in 1977 and 1979, respectively, and the Ph.D.
degree in computer engineering from the University of Massachusetts at Amherst, in
1988. Currently he is a Professor of the School of Information and Communication
Engineering, Sungkyunkwan University, Suwon, Korea, and the Director of the
Ubiquitous Computing Technology Research Institute. His research interests include
distributed and ubiquitous computing, system software and middleware, and
RFID/USN.
Ung-Mo Kim received the B.E. degree in Mathematics from Sungkyunkwan
University, Korea, in 1981 and the M.S. degree in Computer Science form Old
Dominion University, U.S.A. in 1986. His Ph.D. degree was received in Computer
Science from Northwestern University, U.S.A., in 1990. Currently he is a full professor
of School of Information and Communication Engineering, Sungkyunkwan University,
Korea. His research interests include data mining, database security, data warehousing,
GIS and big data

Efficient processing of spatial range queries on wireless broadcast streams

Recomendados

Recomendados

Más contenido relacionado

La actualidad más candente

La actualidad más candente (17)

Destacado

Destacado (9)

Similar a Efficient processing of spatial range queries on wireless broadcast streams

Similar a Efficient processing of spatial range queries on wireless broadcast streams (20)

Último

Último (20)

Efficient processing of spatial range queries on wireless broadcast streams